A power grid operation and maintenance multi-task planning method and system based on space-time features

Abstract

The application relates to the technical field of power grids and discloses a power grid operation and maintenance multi-task planning method and system based on space-time characteristics. The method comprises the following steps: based on equipment time sequence dynamic characteristics and equipment space correlation characteristics, performing power grid operation monitoring tasks and equipment state monitoring tasks to obtain fault detection statistics and fault positioning statistics; performing a risk research and judgment process on the fault detection statistics and the fault positioning statistics and time sequence qualitative trend characteristics to obtain an operation risk assessment result of a target power grid; inputting the equipment space correlation characteristics into each graph neural network for learning to obtain global topology aggregation characteristics of the target power grid and node space characteristics corresponding to each equipment; and inputting the global topology aggregation characteristics and each node space characteristic into a corresponding strategy network for processing to output an operation and maintenance coordination planning strategy of the target power grid. The application not only improves the overall execution efficiency of power grid operation and maintenance work, but also reduces the comprehensive cost of operation and maintenance scheduling.

Description

Technical Field

[0001] This invention relates to the field of power grid technology, and in particular to a multi-task planning method and system for power grid operation and maintenance based on spatiotemporal characteristics. Background Technology

[0002] Traditional power grid operation and maintenance methods rely heavily on periodic inspections and reactive repairs. This model is not only inefficient but also ill-equipped to handle sudden faults, easily leading to large-scale power outages. Therefore, achieving intelligent, refined, and proactive management of power grid operation and maintenance has become a critical issue that the power industry urgently needs to address.

[0003] In recent years, the rapid development of advanced information technologies such as the Internet of Things, big data, cloud computing, and artificial intelligence has provided solid technical support for the intelligent transformation of power grid operation and maintenance. By integrating these technologies, real-time monitoring of power grid equipment status, early warning of faults, and optimized allocation of operation and maintenance resources can be achieved, thereby improving the efficiency and reliability of power grid operation and maintenance. However, as a complex spatiotemporal network, the power grid system's operation and maintenance data has characteristics such as high dimensionality, nonlinearity, and spatiotemporal coupling. How to extract effective features from massive amounts of data to achieve accurate status assessment and fault prediction remains a major technical challenge.

[0004] In existing technologies, feature extraction methods for power grid operation and maintenance mostly focus on a single time or spatial dimension. For example, while time-domain analysis methods can effectively capture the dynamic changes in equipment status, they are difficult to reveal the spatial propagation patterns of faults; while spatial-domain analysis methods focus on the topological relationships between equipment, ignoring the continuity and trends in the time dimension. Due to the lack of in-depth analysis and comprehensive consideration of the complex relationships between multiple tasks, the planning results are difficult to achieve optimal performance. Summary of the Invention

[0005] To address the aforementioned problems in existing technologies, this invention provides a multi-task planning method and system for power grid operation and maintenance based on spatiotemporal characteristics.

[0006] In a first aspect, embodiments of the present invention provide a multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics, including: Acquire multi-source monitoring data covering target power grid equipment, and extract spatiotemporal features from the multi-source monitoring data to obtain equipment temporal dynamic features and equipment spatial correlation features; Based on the temporal dynamic characteristics and spatial correlation characteristics of the equipment, power grid operation monitoring tasks and equipment status monitoring tasks are executed accordingly to obtain fault detection statistics and fault location statistics. A risk assessment process is performed on the fault detection statistics, the fault location statistics, and the time-series qualitative trend characteristics generated by analyzing the fault detection statistics to obtain the operational risk assessment results of the target power grid. Construct a graph neural network and a policy network corresponding to each device in the target power grid, input the spatial association features of the device into each graph neural network for learning, and obtain the global topology aggregation features of the target power grid and the node spatial features corresponding to each device; The global topology aggregation features and the spatial features of each node are input into the corresponding policy network to evaluate the importance of node operation and maintenance and to allocate the probability of operation and maintenance sub-tasks based on the operation risk assessment results, thereby outputting the operation and maintenance collaborative planning strategy for the target power grid.

[0007] Preferably, the step of acquiring multi-source monitoring data covering the target power grid equipment and extracting spatiotemporal features from the multi-source monitoring data to obtain equipment temporal dynamic features and equipment spatial correlation features includes: Access the runtime sequence data and spatial topology data of the target power grid equipment to form multi-source monitoring data; Based on canonical correlation analysis, the time features of the runtime sequence data in the multi-source monitoring data are extracted to obtain the device time-series dynamic features. Spatial features are extracted from the spatial topology data in the multi-source monitoring data by convolutional unsupervised feature learning to obtain the spatial association features of the equipment.

[0008] Preferably, the step of performing power grid operation monitoring tasks and equipment status monitoring tasks based on the temporal dynamic characteristics and spatial correlation characteristics of the equipment to obtain fault detection statistics and fault location statistics includes: The time-series dynamic characteristics of the device are distributed to edge computing nodes to perform power grid operation monitoring tasks, thereby obtaining fault detection statistics. The device spatial association features are deployed to the edge terminal to perform device status monitoring tasks, and fault location statistics are obtained.

[0009] Preferably, the step of distributing the device's time-series dynamic features to edge computing nodes to perform power grid operation monitoring tasks and obtain fault detection statistics includes: The device's time-series dynamic characteristics are distributed to edge computing nodes deployed at the power grid regional operation and maintenance management level; The edge computing node processes the real-time power grid operation data and the time-series dynamic characteristics of the equipment transmitted back by the edge terminal based on Wasserstein distance to generate the operation performance monitoring statistics of the target power grid. The kernel density estimation method is used to determine the operating performance monitoring and control thresholds of the target power grid; By comparing the operational performance monitoring statistics and the operational performance monitoring control threshold, a fault detection statistics that includes the operating conditions of the target power grid are obtained.

[0010] Preferably, the step of deploying the device spatial association features to the edge terminal to perform device status monitoring tasks and obtain fault location statistics includes: Deploy the device spatial association features to the edge terminals installed on each device; The normal operating condition space features and online operating space features of the corresponding device are extracted by using a time slicing strategy for each edge terminal. The normal operating condition spatial characteristics and the online operating spatial characteristics of each device are processed based on Wasserstein distance to generate the corresponding device status monitoring statistics. By combining the fault alarms triggered by the edge computing node, the fault location is obtained through the status monitoring statistics, which includes the faulty device and the fault location.

[0011] Preferably, the risk assessment process performed on the fault detection statistics, the fault location statistics, and the time-series qualitative trend features generated by analyzing the fault detection statistics to obtain the operational risk assessment results of the target power grid includes: Based on the degree of closeness between the fault detection statistics and the operating performance monitoring and control threshold of the target power grid, an operating condition deviation index is constructed. A sliding window is used to perform qualitative trend analysis on the fault detection statistics to generate time-series qualitative trend features; The time-series qualitative trend characteristics and the operating condition deviation index are fused and calculated to obtain the operation risk assessment index of the target power grid; The operational risk assessment indicators are corrected by combining the fault location statistics to obtain the operational risk assessment results of the target power grid.

[0012] Preferably, the step of constructing a graph neural network and a policy network corresponding to each device in the target power grid, and inputting the spatial association features of the devices into each graph neural network for learning, to obtain the global topology aggregation features of the target power grid and the node spatial features corresponding to each device, includes: By uniformly defining the network architecture paradigm and parameter layout rules, a graph neural network and a policy network corresponding to each device in the target power grid are constructed. The structural parameters of all graph neural networks are shared from the same source, and the structure of each policy network is the same but the parameters are different. The spatial association features of the devices are input into each of the graph neural networks to learn the topological relationships, thereby obtaining the global topological aggregation features of the target power grid and the node spatial features corresponding to each device.

[0013] Preferably, the step of inputting the global topology aggregation features and the spatial features of each node into the corresponding policy network to perform node operation and maintenance importance assessment and operation and maintenance sub-task probability allocation based on the operation risk assessment results, and outputting the operation and maintenance collaborative planning strategy for the target power grid, includes: The global topology aggregation feature and the spatial feature of each node are input into the corresponding policy network; For each device, the importance of node operation and maintenance is assessed to obtain a quantitative value of the importance of node operation and maintenance for the corresponding device; Based on the operational risk assessment results, the quantitative value of the operational importance of each node is adjusted using a probabilistic weighting method to obtain the operational subtask allocation probability for each device. By integrating the allocation probabilities of all the operation and maintenance sub-tasks, and solving for the optimal operation and maintenance sub-task allocation scheme, an operation and maintenance collaborative planning strategy for the target power grid is generated.

[0014] Preferably, the step of assessing the node maintenance importance of each device to obtain a quantitative value for the node maintenance importance of the corresponding device includes: Based on the global topology aggregation features and the spatial features of each node, an attention mechanism is used to evaluate the node operation and maintenance importance of each device, thereby obtaining a quantitative value of the node operation and maintenance importance of the corresponding device.

[0015] Secondly, embodiments of the present invention provide a power grid operation and maintenance multi-task planning system based on spatiotemporal characteristics, including: The feature extraction module is used to acquire multi-source monitoring data covering the target power grid equipment, and to extract spatiotemporal features from the multi-source monitoring data to obtain the equipment temporal dynamic features and equipment spatial correlation features; The task execution module is used to execute power grid operation monitoring tasks and equipment status monitoring tasks based on the device's temporal dynamic characteristics and spatial correlation characteristics, and to obtain fault detection statistics and fault location statistics. The risk assessment module is used to perform a risk assessment process on the fault detection statistics, the fault location statistics, and the time-series qualitative trend characteristics generated by analyzing the fault detection statistics, so as to obtain the operational risk assessment results of the target power grid. A learning module is constructed to build a graph neural network and a policy network corresponding to each device in the target power grid. The spatial association features of the devices are input into each graph neural network for learning, so as to obtain the global topology aggregation features of the target power grid and the node spatial features corresponding to each device. The strategy output module is used to input the global topology aggregation features and the spatial features of each node into the corresponding strategy network to perform node operation and maintenance importance assessment and operation and maintenance sub-task probability allocation based on the operation risk assessment results, and output the operation and maintenance collaborative planning strategy of the target power grid.

[0016] Compared with existing technologies, the multi-task planning method and system for power grid operation and maintenance based on spatiotemporal characteristics proposed in this invention have the following advantages at least one point: First, by jointly extracting the temporal dynamic features and spatial correlation features of equipment, we can capture the evolutionary patterns of power grid equipment operation and the topological relationships between equipment. Based on unsupervised representation and correlation analysis techniques, we can delve into the implicit information inherent in the data, overcome the limitations of single feature analysis, and improve the comprehensiveness and accuracy of power grid equipment operation status perception.

[0017] Secondly, the overall power grid operation fault detection and individual equipment fault location are completed separately. Then, combined with time-series qualitative trend analysis and the deviation of operating conditions indicators, risk assessment is carried out. This can quickly identify real-time operational anomalies, accurately locate faults, and predict the development and evolution of faults, thus realizing the transformation of power grid faults from post-event handling to pre-event early warning and effectively curbing the expansion of fault scope.

[0018] Finally, by learning the grid topology association information through graph neural networks, we obtain the global topology aggregation features and node spatial features. Then, through the attention mechanism, we complete the quantitative assessment of the importance of node operation and maintenance. Combined with the grid operation risk assessment results, we realize the probabilistic and reasonable allocation of operation and maintenance sub-tasks, complete the collaborative planning of operation and maintenance tasks, improve the overall execution efficiency of grid operation and maintenance work, and reduce the comprehensive cost of operation and maintenance scheduling. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the process of obtaining fault detection statistics and fault location statistics according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the process for obtaining the operation and maintenance collaborative planning strategy according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a power grid operation and maintenance multi-task planning system based on spatiotemporal characteristics according to an embodiment of the present invention; Figure label: 01. Feature Extraction Module; 02. Task Execution Module; 03. Risk Assessment Module; 04. Learning Construction Module; 05. Strategy Output Module. Detailed Implementation

[0020] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0021] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by those skilled in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0022] like Figure 1 The diagram shown is a flowchart illustrating a multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics, according to an embodiment of the present invention. (Refer to...) Figure 1 This invention provides a multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics, comprising the following steps: S1. Acquire multi-source monitoring data covering the target power grid equipment, and extract spatiotemporal features from the multi-source monitoring data to obtain the equipment temporal dynamic features and equipment spatial correlation features; Specifically, step S1 includes: 1) Access the runtime sequence data and spatial topology data of the target power grid equipment to form multi-source monitoring data; Multi-source monitoring data includes runtime sequence data and spatial topology data.

[0023] Specifically, runtime sequence data is taken from the power grid SCADA system, synchronous phasor measurement devices, and equipment sensors, including continuous time-series monitoring indicators such as equipment voltage, current, active power, reactive power, operating temperature, and vibration frequency, with sampling frequencies adapted to the dynamic monitoring needs of the power grid. Spatial topology data is taken from the power grid GIS system, equipment ledger, and topology connection database, including topology information such as equipment node numbers, line connection relationships, equipment adjacency matrices, and topology distances, comprehensively representing the physical connections and spatial coupling relationships between power grid equipment.

[0024] 2) Based on canonical correlation analysis, time features are extracted from the runtime sequence data in the multi-source monitoring data to obtain the dynamic time-series features of the equipment; Kernel Canonical Correlation Analysis (KCCA) aims to project high-dimensional and nonlinear runtime data into a high-dimensional feature space through kernel mapping, thereby maximizing the correlation between time-series variables and performance indicators and effectively extracting nonlinear time-series features.

[0025] The process of extracting time features is explained in detail below: ① The Gaussian kernel function is selected as the kernel mapping operator, and the formula is: in, Indicates the first The time series sample and the first Kernel similarity between time series samples and Represents time series samples at different times. This represents the kernel bandwidth, balancing feature smoothness and discriminability.

[0026] ② Construct the autovariance matrix of time series data Performance index autovariance matrix And the time-performance covariance matrix ; ③ Constructing a matrix The formula is: Solving for the optimal projection vector using singular value decomposition and To maximize the correlation coefficient The formula is: ④ Screening based on cumulative correlation contribution rate ≥ 90% A set of typical related features are used to map the high-dimensional temporal features after projection into low-dimensional temporal dynamic features, thus fully preserving the evolution of equipment performance over time.

[0027] 3) Spatial features are extracted from the spatial topology data in the multi-source monitoring data through convolutional unsupervised feature learning to obtain the spatial association features of the equipment.

[0028] Convolutional unsupervised feature learning is based on convolutional canonical correlation analysis, which eliminates the need for manual labeling and unsupervised mining of device associations and fault propagation features hidden in spatial topology data.

[0029] The process of spatial feature extraction is explained in detail below: First, adopt the same time slice length as the time series data. Constructing a spatial data matrix , This indicates the number of spatial measurement points, with each row of the matrix corresponding to the time series data of a single measurement point.

[0030] Secondly, construct the cross-correlation enhancement matrix. Introducing the order of cross-correlation It integrates the correlation information of neighboring measurement points and strengthens the spatial dimension coupling characteristics.

[0031] Furthermore, a two-layer convolutional canonical correlation analysis network is constructed. The first convolutional layer extracts local spatial correlation features, and the second convolutional layer fuses global topological dependencies. Each layer extracts spatial features by maximizing canonical correlation, and the number of features is selected based on a cumulative contribution rate of ≥90%.

[0032] Finally, the convolutional output features are reconstructed and mapped to obtain the dimensionality-reduced device spatial association features, which accurately characterize the topological connection strength, spatial neighborhood dependency, and fault spatial propagation path between devices.

[0033] S2. Based on the temporal dynamic characteristics and spatial correlation characteristics of the equipment, perform power grid operation monitoring tasks and equipment status monitoring tasks to obtain fault detection statistics and fault location statistics. Based on the temporal dynamic characteristics and spatial correlation characteristics of devices, monitoring tasks are executed in a layered manner relying on the cloud-edge-device collaborative architecture.

[0034] Specifically, the temporal dynamic characteristics of the equipment are deployed to the edge side to focus on the dynamic monitoring of the overall operation of the power grid, while the spatial correlation characteristics of the equipment are deployed to the terminal side to focus on the local state perception of a single device. The two types of tasks are carried out in parallel, capturing operational anomalies from the time dimension and locking fault locations from the spatial dimension, respectively. Finally, fault detection statistics and fault location statistics are output synchronously, realizing dual-dimensional monitoring of the overall power grid anomaly identification and local fault location.

[0035] like Figure 2 As shown, this is a flowchart illustrating step S2. (Refer to...) Figure 2 Step S2 includes: S201. Distribute the device's time-series dynamic characteristics to edge computing nodes to perform power grid operation monitoring tasks and obtain fault detection statistics; Specifically, step S201 includes: 1) Distribute the device's time-series dynamic characteristics to edge computing nodes deployed at the power grid area operation and maintenance management level; The power grid regional operation and maintenance management level corresponds to the comprehensive management and control L2 level. Edge computing nodes are deployed in regional substations and core nodes of operation and maintenance zones, possessing low-latency real-time computing power. The equipment time-series dynamic characteristics are output by the cloud-side time feature extraction model (core canonical correlation analysis model) and distributed to the edge computing nodes via encrypted dedicated communication links. This feature has integrated the time-series information associated with key performance indicators of power grid operation and projection matrix parameters, providing standardized time-series input for real-time edge monitoring and ensuring cross-node data consistency and transmission security.

[0036] 2) The real-time operation data of the power grid and the time-series dynamic characteristics of the equipment transmitted back by the edge terminal are processed based on Wasserstein distance through the edge computing nodes to generate the operation performance monitoring statistics of the target power grid; Wasserstein distance (WD, bulldozer distance) is used to quantify the optimal transmission difference between two time series distributions, adapting to the high-dimensional nonlinear characteristics of power grid data. Edge terminals collect and transmit real-time operational data such as voltage, current, and active power from the target power grid. Edge computing nodes fuse the real-time operational data (online time series distribution) with the dynamic time series characteristics of the equipment (reference time series distribution) to construct time series sample pairs. and According to the WD formula: Constraints: in, This represents the statistical quantity for operational performance monitoring. Represents the baseline time series distribution weights. Indicates the online time-series distribution weights. Represents the L2 norm distance between time series samples. , This represents the optimal transmission probability weight.

[0037] After constraint optimization, operational performance monitoring statistics reflecting the degree of deviation of power grid operation are generated.

[0038] 3) Kernel density estimation is used to determine the target power grid's operational performance monitoring and control thresholds; Kernel density estimation (KDE) is a nonparametric probability density fitting method that does not require pre-setting the distribution type and adapts to the complexity of the time-series distribution of normal power grid operation. Using historical operational performance monitoring statistics under normal operating conditions as samples, a Gaussian kernel function is used to fit the probability density curve. To balance the risks of missed detections and false detections, a 90%–99% confidence interval is preferred. In this embodiment, the 95% confidence quantile is set as the operational performance monitoring control threshold. This threshold serves as the criterion for determining normal / fault operating conditions, representing the maximum permissible operational deviation limit of the power grid.

[0039] 4) Compare the operating performance monitoring statistics and the operating performance monitoring control thresholds to obtain fault detection statistics that include the operating conditions of the target power grid.

[0040] Comparison of operating conditions based on binary judgment rules: When running performance monitoring statistics When the power grid is in a fault condition, the power grid is determined to be in a fault condition; when the operating performance monitoring statistics are... At that time, it is determined that the power grid is in normal operating condition.

[0041] The fault detection statistics synchronously include three types of information: operating condition judgment results, deviation value, and anomaly occurrence time sequence, providing time-series anomaly basis for cloud-side risk assessment.

[0042] S202. Deploy the spatial association features of the equipment to the edge terminal to perform equipment status monitoring tasks and obtain fault location statistics.

[0043] Specifically, step S202 includes: 1) Deploy device spatial association features to edge terminals installed on each device; Edge terminals are installed one-to-one on target power grid devices such as transformers, circuit breakers, and lines, and have local data acquisition and lightweight computing capabilities. The spatial correlation features of the devices are output by the cloud-side spatial feature extraction model (convolutional unsupervised feature learning model) and deployed to the local storage of the corresponding edge terminal, adapting to the specific status monitoring needs of a single device and avoiding cross-device feature interference.

[0044] 2) Extract the normal operating space features and online operating space features of the corresponding device by using a time slicing strategy for each edge terminal; The time-slicing strategy divides the equipment spatial monitoring time series into fixed durations (preferably 10-30 minutes). During the offline steady-state period, spatial features under normal operating conditions are extracted; these features reflect the topological correlation baseline distribution when the equipment is fault-free. During the online real-time period, online operating spatial features are extracted; these features characterize the current spatial coupling dynamics of the equipment. Both types of features are output through unsupervised learning of the spatial feature extraction model, accurately capturing the differences in the local spatial state of the equipment.

[0045] 3) Process the normal operating space characteristics and online operating space characteristics of each device based on Wasserstein distance to generate the corresponding device status monitoring statistics; Using the Wasserstein distance to quantify the distributional differences between two types of spatial features, spatial sample pairs are constructed. (Normal spatial characteristics) and (Online spatial characteristics), substituting into the WD formula: in, This represents the equipment status monitoring statistics, and the calculation formulas for the other variables are defined in the same way as those for the operational performance monitoring statistics. This statistic directly quantifies the degree of deviation between the current spatial state of the equipment and the normal baseline; the larger the value, the higher the probability of equipment abnormality.

[0046] 4) Combine the fault alarms triggered by the edge computing nodes and use the status monitoring statistics to locate the fault, and obtain the fault location statistics that include the faulty equipment and the fault location.

[0047] After an edge computing node triggers a fault alarm, it synchronously sends a location command to all edge terminals. Each edge terminal, based on local device status monitoring statistics, filters target devices whose deviation exceeds the spatial status monitoring control threshold determined by kernel density estimation. Combining the unique number in the device ledger, installation location, and topology connection relationship, it locates the faulty device and fault location. The final output fault location statistics include the faulty device ID, geographic coordinates, topology associated nodes, and spatial deviation value, providing a spatial dimension basis for operation and maintenance resource scheduling.

[0048] S3. Perform a risk assessment process on the fault detection statistics, fault location statistics, and time-series qualitative trend characteristics generated by analyzing the fault detection statistics to obtain the operational risk assessment results of the target power grid. Specifically, step S3 includes: 1) Construct an index of operating condition deviation based on the proximity between fault detection statistics and the target power grid's operating performance monitoring and control thresholds; The deviation from operating conditions index uses the PCL index to quantify the relative deviation between fault detection statistics and operational performance monitoring and control thresholds, reflecting the difference between the current operating state of the power grid and the normal baseline. Its calculation formula is: in, This index represents the degree of deviation from the operating condition, and its value range is... A smaller value indicates that the power grid is operating closer to normal conditions and the risk is lower; a larger value indicates that the deviation from normal conditions is more significant and the risk of failure is higher.

[0049] 2) A sliding window is used to perform qualitative trend analysis on fault detection statistics to generate time-series qualitative trend features; Qualitative trend analysis using a sliding window (QTA) is employed, with a fixed-length sliding window (window length...). (Preferably 5 to 20 time-series sampling points) to perform segment-by-segment trend analysis on the fault detection statistics in the form of time-series sequences.

[0050] By identifying seven standard trend primitives (constant A, strongly increasing B, moderately increasing C, weak D, weakly decreasing E, moderately decreasing F, and strongly decreasing G) through derivative feature matching, information on trend direction, rate of change, and duration is extracted to generate time-series qualitative trend features. These features quantify the dynamic evolution of fault detection statistics, compensating for the inability to predict risk evolution based solely on static deviation values, and supplementing risk assessment with trend dimension information.

[0051] 3) The time-series qualitative trend characteristics and the deviation of operating conditions are integrated and calculated to obtain the operation risk assessment index of the target power grid; A weighted fusion of time-series qualitative trend characteristics and operating condition deviation indicators is achieved based on a trend fusion function.

[0052] Specifically, the trend fusion weights are calculated: in, Indicates trend fusion weight, Indicates trend contribution rate (matching primitive) The corresponding quantization value has a range of values. , Indicators representing trend continuity reflect the impact of the duration of a trend. , Indicates the deviation correlation coefficient. .

[0053] Furthermore, construct comprehensive operational risk assessment indicators: in, It represents the operational risk assessment indicators, integrating quantitative deviation and qualitative evolution trends. Under normal operating conditions, it reflects the probability of potential failures, and under fault conditions, it reflects the risk of deterioration, thus achieving multi-dimensional risk quantification.

[0054] 4) The operational risk assessment indicators are corrected by combining the fault location statistics to obtain the operational risk assessment results of the target power grid.

[0055] Based on information such as faulty device ID, topological associated nodes, and spatial deviation value included in the fault location statistics, the operational risk assessment indicators are weighted and corrected from two dimensions: equipment importance and fault impact range.

[0056] Specifically, faults corresponding to core equipment and critical nodes in the main grid are assigned higher correction weights; faults in non-critical equipment at the end are assigned appropriately lower weights to accurately distinguish the actual hazard levels under the same risk indicators. The corrected target power grid operation risk assessment results include risk level, risk value, fault impact range, and evolution trend prediction, providing a graded basis for operation and maintenance decisions.

[0057] S4. Construct a graph neural network and a policy network corresponding to each device in the target power grid. Input the spatial association features of the devices into each graph neural network for learning to obtain the global topology aggregation features of the target power grid and the node spatial features corresponding to each device. Specifically, step S4 includes: 1) By uniformly defining the network architecture paradigm and parameter deployment rules, construct the graph neural network and policy network corresponding to each device in the target power grid; All graph neural networks share the same structural parameters, and each policy network has the same structure but different parameters.

[0058] The graph neural network adopts a unified message passing architecture, which includes four major modules: message creation, message collection, message passing, and graph embedding. The graph neural networks corresponding to all devices share core parameters such as network weights, activation functions, neighbor aggregation functions, and global kernels, ensuring the consistency and universality of power grid topology feature learning. The policy network adopts a parallel sub-network architecture, with each device corresponding to an independent sub-network. All policy networks maintain the same structure in terms of layers, dimensions, activation functions, etc., but the neuron parameters are initialized independently to adapt to the exclusive decision logic of a single device.

[0059] Graph neural networks focus on mining topological correlation features in power grids, while policy networks are responsible for calculating the importance of node operation and maintenance. Together, they construct a basic model architecture that adapts to multi-task planning in power grids.

[0060] 2) Input the spatial association features of the equipment into each graph neural network to learn the topological relationship, and obtain the global topological aggregation features of the target power grid and the node spatial features corresponding to each equipment.

[0061] The spatial association features of the device include information such as device node attributes, adjacency connections, and spatial coupling associations. After inputting into the graph neural network, it performs multi-round iterative topology message passing.

[0062] No. The formula for updating the features of round nodes is: in, Represents a node No. Wheel features, This represents the activation function. Represents a node Neighbors gather, This indicates the prediction results for edge importance. Indicates the global kernel. Representing neighboring nodes No. Wheel features.

[0063] through Round iteration ( After completing multi-hop neighbor information fusion (preferably 3 to 5 rounds), global topology aggregation features are output through global pooling to characterize the overall topology structure and global correlation patterns of the power grid. At the same time, the spatial features of each device's dedicated nodes are output to reflect the topology neighborhood attributes and spatial correlation characteristics of a single device, providing standardized feature inputs for subsequent policy network importance assessment.

[0064] S5. Input the global topology aggregation features and the spatial features of each node into the corresponding strategy network to evaluate the importance of node operation and maintenance and to allocate the probability of operation and maintenance sub-tasks based on the results of the operation risk assessment. Output the operation and maintenance collaborative planning strategy for the target power grid.

[0065] Specifically, based on the global topology aggregation features output by the graph neural network and the spatial features of each device node, and relying on the parallelized strategy network, the probability allocation of sub-tasks such as attention-driven node operation and maintenance importance quantification and risk fusion is completed in sequence. Finally, the optimal solution is solved to realize the collaborative allocation of power grid operation and maintenance resources, and outputs an operation and maintenance collaborative planning strategy that takes into account both topology importance and operational risks, supporting the efficient coordination of multiple tasks in the power grid.

[0066] like Figure 3 As shown, this is a flowchart illustrating step S5. (Refer to...) Figure 3 Step S5 includes: S501. Input the global topology aggregation features and the spatial features of each node into the corresponding policy network; The global topology aggregation features and corresponding device node spatial features are input one-to-one into each independent policy network. The global topology aggregation features represent the overall topology association information of the power grid, providing a global decision-making basis for the policy network. The node spatial features carry the unique topology neighborhood attributes of a single device, ensuring the targeted nature of device-level decisions. The parallel input of these two types of features, with dimensions adapted to the policy network input layer, ensures the collaborative integration of global and local information, providing complete feature support for subsequent importance assessment.

[0067] S502. Assess the importance of node operation and maintenance for each device to obtain a quantitative value of the importance of node operation and maintenance for the corresponding device. Based on the global topology aggregation features and the spatial features of each node, and combined with the attention mechanism, the node operation and maintenance importance of each device is evaluated, and the quantitative value of the node operation and maintenance importance of the corresponding device is obtained.

[0068] Specifically, the policy network employs a two-stage attention mechanism to quantify the importance of node operation and maintenance: The first step is to construct a graph content embedding vector by splicing the global topology aggregation features and the device node spatial features. ; The second step is to generate query vectors respectively. Key vector Value vector The formula is: in, , , Represents network parameters, Represents a node Spatial characteristics.

[0069] The third step is to calculate compatibility: in, This indicates the feature compatibility between the query vector and the key vector. This represents the scaling factor for the key vector dimension. The attention weights are output after softmax. .

[0070] The fourth step is to obtain the device feature vector by weighted summation. : Then, the tanh activation mapping outputs a quantitative value of the node's operational importance. Quantization range A higher value indicates a higher priority for equipment maintenance.

[0071] S503. Based on the operational risk assessment results, the quantitative value of the operational importance of each node is adjusted by probability weighting to obtain the probability of operational subtask allocation for each device. Based on the results of the operational risk assessment As a dynamic weighting, the quantitative values ​​of node operation and maintenance importance are probabilistically weighted, and the weighting formula is as follows: in, This represents the probability of assigning maintenance subtasks. This represents the operational risk assessment result for the i-th device.

[0072] Specifically, high-risk and high-importance equipment is assigned a higher allocation probability, achieving a two-dimensional coupling of risk and importance, avoiding decision-making bias based on a single dimension, and determining the probability value. The sum is normalized to satisfy the probability distribution characteristics.

[0073] S504. Integrate the allocation probabilities of all operation and maintenance sub-tasks, and generate the operation and maintenance collaborative planning strategy for the target power grid by solving the optimal operation and maintenance sub-task allocation scheme.

[0074] Integrate the allocation probabilities of all equipment operation and maintenance subtasks, and construct the objective optimization function based on the OR-Tools optimization tool, combining policy gradient and advantage function: in, This represents the optimal parameters of the policy network. Indicates policy network parameters, Represents the training set, Indicates the expected value of the training set. Represents the policy distribution. Represents the dominance function. Indicates the first Subtask allocation scheme for secondary sampling. This indicates the approximate number of samples.

[0075] The goal is to maximize operational benefits and minimize resource losses, and to output the optimal allocation scheme for subtasks of each device. Finally, by integrating the decision results of each device, a target power grid operation and maintenance collaborative planning strategy is generated, which includes operation and maintenance priorities, resource allocation and task division, to achieve collaborative operation and maintenance of multiple devices and optimal resource allocation.

[0076] This invention discloses a multi-task planning method for power grid operation and maintenance based on spatiotemporal features. By jointly extracting the temporal dynamic features and spatial correlation features of equipment, it captures the evolutionary patterns of power grid equipment operation and the topological relationships between equipment. Based on unsupervised representation and correlation analysis techniques, it deeply mines the inherent implicit information in the data, overcoming the limitations of single feature analysis and improving the comprehensiveness and accuracy of power grid equipment operation status perception. It separately completes the detection of overall power grid operation faults and the location of individual equipment faults. Then, it combines temporal qualitative trend analysis with operating condition deviation indicators to perform risk assessment. This method can quickly identify real-time operational anomalies, accurately locate faults, and predict the development and evolution trends of faults, realizing the transformation of power grid fault handling from post-event handling to pre-event warning, effectively curbing the expansion of fault scope. By learning the power grid topology correlation information through graph neural networks, it obtains global topology aggregation features and node spatial features. Then, it completes the quantitative assessment of the importance of node operation and maintenance through an attention mechanism. Combined with the power grid operation risk assessment results, it realizes the probabilistic and reasonable allocation of operation and maintenance sub-tasks, completes the collaborative planning of operation and maintenance tasks, improves the overall execution efficiency of power grid operation and maintenance work, and reduces the comprehensive cost of operation and maintenance scheduling.

[0077] like Figure 4 The diagram shown is a structural schematic of a multi-task planning system for power grid operation and maintenance based on spatiotemporal characteristics, according to an embodiment of the present invention. (Refer to...) Figure 4 An embodiment of the present invention provides a multi-task planning system for power grid operation and maintenance based on spatiotemporal characteristics, comprising: Feature extraction module 01 is used to acquire multi-source monitoring data covering target power grid equipment, and to extract spatiotemporal features from the multi-source monitoring data to obtain equipment temporal dynamic features and equipment spatial correlation features; Task execution module 02 is used to execute power grid operation monitoring tasks and equipment status monitoring tasks based on the equipment time-series dynamic characteristics and equipment spatial correlation characteristics, and obtain fault detection statistics and fault location statistics. Risk assessment module 03 is used to perform risk assessment process on fault detection statistics, fault location statistics and time series qualitative trend characteristics generated by analyzing fault detection statistics, and obtain the operation risk assessment results of the target power grid. The learning module 04 is constructed to build the graph neural network and policy network corresponding to each device in the target power grid. The spatial association features of the devices are input into each graph neural network for learning, so as to obtain the global topology aggregation features of the target power grid and the node spatial features corresponding to each device. The strategy output module 05 is used to input the global topology aggregation features and the spatial features of each node into the corresponding strategy network to evaluate the importance of node operation and maintenance and to allocate the probability of operation and maintenance sub-tasks based on the results of the operation and maintenance risk assessment, and output the operation and maintenance collaborative planning strategy of the target power grid.

[0078] It should be noted that the modules in the aforementioned spatiotemporal feature-based multi-task planning system for power grid operation and maintenance can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module. For specific limitations regarding the spatiotemporal feature-based multi-task planning system for power grid operation and maintenance, please refer to the limitations of the spatiotemporal feature-based multi-task planning method for power grid operation and maintenance described above; both have the same function and role, and will not be repeated here.

[0079] In summary, this invention provides a multi-task planning method and system for power grid operation and maintenance based on spatiotemporal features. By jointly extracting the temporal dynamic features and spatial correlation features of equipment, it captures the evolutionary patterns of power grid equipment operation and the topological relationships between equipment. Based on unsupervised representation and correlation analysis techniques, it deeply mines the inherent implicit information in the data, overcoming the limitations of single feature analysis and improving the comprehensiveness and accuracy of power grid equipment operation status perception. It separately completes the detection of overall power grid operation faults and the location of individual equipment faults. Then, by combining temporal qualitative trend analysis with the deviation index of operating conditions, it performs risk assessment, which can quickly identify real-time operational anomalies, accurately locate faults, and predict the development and evolution trend of faults, realizing the transformation of power grid faults from post-event handling to pre-event warning, effectively curbing the expansion of fault scope. By learning the power grid topology correlation information through graph neural networks, it obtains the global topology aggregation features and node spatial features. Then, through the attention mechanism, it completes the quantitative assessment of the importance of node operation and maintenance. Combined with the power grid operation risk assessment results, it realizes the probabilistic and reasonable allocation of operation and maintenance sub-tasks, completes the collaborative planning of operation and maintenance tasks, improves the overall execution efficiency of power grid operation and maintenance work, and reduces the comprehensive cost of operation and maintenance scheduling.

[0080] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0081] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics, characterized in that, include: Acquire multi-source monitoring data covering target power grid equipment, and extract spatiotemporal features from the multi-source monitoring data to obtain equipment temporal dynamic features and equipment spatial correlation features; Based on the temporal dynamic characteristics and spatial correlation characteristics of the equipment, power grid operation monitoring tasks and equipment status monitoring tasks are executed accordingly to obtain fault detection statistics and fault location statistics. A risk assessment process is performed on the fault detection statistics, the fault location statistics, and the time-series qualitative trend characteristics generated by analyzing the fault detection statistics to obtain the operational risk assessment results of the target power grid. Construct a graph neural network and a policy network corresponding to each device in the target power grid, input the spatial association features of the device into each graph neural network for learning, and obtain the global topology aggregation features of the target power grid and the node spatial features corresponding to each device; The global topology aggregation features and the spatial features of each node are input into the corresponding policy network to evaluate the importance of node operation and maintenance and to allocate the probability of operation and maintenance sub-tasks based on the operation risk assessment results, thereby outputting the operation and maintenance collaborative planning strategy for the target power grid.

2. The multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to claim 1, characterized in that, The process of acquiring multi-source monitoring data covering the target power grid equipment and extracting spatiotemporal features from the multi-source monitoring data to obtain equipment temporal dynamic features and equipment spatial correlation features includes: Access the runtime sequence data and spatial topology data of the target power grid equipment to form multi-source monitoring data; Based on canonical correlation analysis, the time features of the runtime sequence data in the multi-source monitoring data are extracted to obtain the device time-series dynamic features. Spatial features are extracted from the spatial topology data in the multi-source monitoring data by convolutional unsupervised feature learning to obtain the spatial association features of the equipment.

3. The multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to claim 1, characterized in that, The process of performing power grid operation monitoring tasks and equipment status monitoring tasks based on the temporal dynamic characteristics and spatial correlation characteristics of the equipment yields fault detection statistics and fault location statistics, including: The time-series dynamic characteristics of the device are distributed to edge computing nodes to perform power grid operation monitoring tasks, thereby obtaining fault detection statistics. The device spatial association features are deployed to the edge terminal to perform device status monitoring tasks, and fault location statistics are obtained.

4. The multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to claim 3, characterized in that, The step of distributing the device's time-series dynamic characteristics to edge computing nodes to perform power grid operation monitoring tasks and obtaining fault detection statistics includes: The device's time-series dynamic characteristics are distributed to edge computing nodes deployed at the power grid regional operation and maintenance management level; The edge computing node processes the real-time power grid operation data and the time-series dynamic characteristics of the equipment transmitted back by the edge terminal based on Wasserstein distance to generate the operation performance monitoring statistics of the target power grid. The kernel density estimation method is used to determine the operating performance monitoring and control thresholds of the target power grid; By comparing the operational performance monitoring statistics and the operational performance monitoring control threshold, a fault detection statistics that includes the operating conditions of the target power grid are obtained.

5. The multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to claim 3, characterized in that, The step of deploying the device spatial association features to the edge terminal to perform device status monitoring tasks and obtain fault location statistics includes: Deploy the device spatial association features to the edge terminals installed on each device; The normal operating condition space features and online operating space features of the corresponding device are extracted by using a time slicing strategy for each edge terminal. The normal operating condition spatial characteristics and the online operating spatial characteristics of each device are processed based on Wasserstein distance to generate the corresponding device status monitoring statistics. By combining the fault alarms triggered by the edge computing node, the fault location is obtained through the status monitoring statistics, which includes the faulty device and the fault location.

6. The multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to claim 1, characterized in that, The risk assessment process is performed on the fault detection statistics, the fault location statistics, and the time-series qualitative trend characteristics generated by analyzing the fault detection statistics to obtain the operational risk assessment results of the target power grid, including: Based on the degree of closeness between the fault detection statistics and the operating performance monitoring and control threshold of the target power grid, an operating condition deviation index is constructed. A sliding window is used to perform qualitative trend analysis on the fault detection statistics to generate time-series qualitative trend features; The time-series qualitative trend characteristics and the operating condition deviation index are fused and calculated to obtain the operation risk assessment index of the target power grid; The operational risk assessment indicators are corrected by combining the fault location statistics to obtain the operational risk assessment results of the target power grid.

7. The multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to claim 1, characterized in that, The process of constructing a graph neural network and a policy network corresponding to each device in the target power grid involves inputting the spatial association features of the devices into each graph neural network for learning, thereby obtaining the global topology aggregation features of the target power grid and the node spatial features corresponding to each device, including: By uniformly defining the network architecture paradigm and parameter layout rules, a graph neural network and a policy network corresponding to each device in the target power grid are constructed. The structural parameters of all graph neural networks are shared from the same source, and the structure of each policy network is the same but the parameters are different. The spatial association features of the devices are input into each of the graph neural networks to learn the topological relationships, thereby obtaining the global topological aggregation features of the target power grid and the node spatial features corresponding to each device.

8. The multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to claim 1, characterized in that, The process of inputting the global topology aggregation features and the spatial features of each node into the corresponding policy network to perform node operation and maintenance importance assessment and operation and maintenance sub-task probability allocation based on the operation risk assessment results, and outputting the operation and maintenance collaborative planning strategy for the target power grid, includes: The global topology aggregation feature and the spatial feature of each node are input into the corresponding policy network; For each device, the importance of node operation and maintenance is assessed to obtain a quantitative value of the importance of node operation and maintenance for the corresponding device; Based on the operational risk assessment results, the quantitative value of the operational importance of each node is adjusted using a probabilistic weighting method to obtain the operational subtask allocation probability for each device. By integrating the allocation probabilities of all the operation and maintenance sub-tasks, and solving for the optimal operation and maintenance sub-task allocation scheme, an operation and maintenance collaborative planning strategy for the target power grid is generated.

9. The multi-task planning method for power grid operation and maintenance based on spatiotemporal characteristics according to claim 8, characterized in that, The process of assessing the importance of node operation and maintenance for each device to obtain a quantitative value for the node operation and maintenance importance of the corresponding device includes: Based on the global topology aggregation features and the spatial features of each node, an attention mechanism is used to evaluate the node operation and maintenance importance of each device, thereby obtaining a quantitative value of the node operation and maintenance importance of the corresponding device.

10. A multi-task planning system for power grid operation and maintenance based on spatiotemporal characteristics, characterized in that, include: The feature extraction module is used to acquire multi-source monitoring data covering the target power grid equipment, and to extract spatiotemporal features from the multi-source monitoring data to obtain the equipment temporal dynamic features and equipment spatial correlation features; The task execution module is used to execute power grid operation monitoring tasks and equipment status monitoring tasks based on the device's temporal dynamic characteristics and spatial correlation characteristics, and to obtain fault detection statistics and fault location statistics. The risk assessment module is used to perform a risk assessment process on the fault detection statistics, the fault location statistics, and the time-series qualitative trend characteristics generated by analyzing the fault detection statistics, so as to obtain the operational risk assessment results of the target power grid. A learning module is constructed to build a graph neural network and a policy network corresponding to each device in the target power grid. The spatial association features of the devices are input into each graph neural network for learning, so as to obtain the global topology aggregation features of the target power grid and the node spatial features corresponding to each device. The strategy output module is used to input the global topology aggregation features and the spatial features of each node into the corresponding strategy network to perform node operation and maintenance importance assessment and operation and maintenance sub-task probability allocation based on the operation risk assessment results, and output the operation and maintenance collaborative planning strategy of the target power grid.

Images